Comment by davydm

This is my experience of all the codegen ais I've seen, and, honestly, it's more efficient for me to just figure the problem out on my own instead of sifting through the slop.

I don't think I'm an outlier either. I think most people using these tools are wasting their time having to debug hallucinated bullshit, instead of learning how to solve the problem themselves, by reading accurate, authoritative resources on the subject, ingesting and compiling that data, and experimenting to make it stick. It just feels like less work when the ai botches it and you have to fix it, vs the mental mountain you have to climb to truly get there, but the journey is overall more efficient when you do take the uncomfortable path.